Download - Northwestern Universityjnd260/pub... · Web viewThe first column of Table 5 shows that as an individual’s ratings for Schakowsky on incumbency, traits, and policy increase, so does

Campaign Rhetoric and the Incumbency Advantage*

James N. Druckman Payson S. Wild Professor of Political Science

Northwestern University601 University PlaceEvanston, IL 60208

[email protected]

Martin J. KiferAssociate Professor of Political Science

High Point University161 Norcross Hall

One University ParkwayHigh Point, NC 27268

[email protected]

Michael Parkin Erwin N. Griswold Professor of Politics

Oberlin College203 Rice Hall

10 North Professor StreetOberlin, OH 44074

[email protected]

*We thank S.R. Gubitz, Adam Howat, and Kumar Ramanathan for research assistance.

1

mailto:[email protected]



Abstract

The congressional incumbency advantage reflects an inequity in competition – candidates receive an electoral edge simply because they hold office. Scholars have identified an array of factors that contribute to the incumbency advantage; however, the role of electoral campaigns has been largely ignored. We argue that campaigns are a mechanism through which the incumbency advantage works. All else constant, incumbents focus their campaigns on factors that reflect their standing position, such as their familiarity to voters and actions taken for their district/state. Voters consequently rely on such incumbency factors when making their decisions. The outcome is challengers are at an extreme disadvantage, and campaigns offer scant substantive engagement. We offer evidence for these dynamics with a large-scale content analysis of campaign websites and an experiment. In so doing, we highlight a challenge to theories of democratic representation that highlight equal competition and/or substantive campaign engagement.

2

Competition is fundamental to democracy because it ensures choice for citizens and

facilitates electoral accountability. Substantial scholarship shows how institutions, particularly

electoral systems, influence the nature of competition and can, at times, privilege certain office-

seekers. One widely discussed manifestation concerns the “incumbency advantage” in United

States congressional elections. This refers to the electoral benefit a candidate receives simply due

to being an incumbent, holding all else constant. Various measures show that the incumbency

advantage has fluctuated between roughly 2% and 12% of votes since the 1950s (e.g., Gelman &

King, 1990; Ansolabehere, Snyder, & Stewart, 2000; Lee, 2008; Fowler & Hall, 2014; Jacobson,

2015).1

How does this advantage affect congressional campaign rhetoric, and how does this

rhetoric influence voters? Surprisingly, these questions have received scant attention, as most

have focused on other aspects of incumbency dynamics such as candidate selection, challenger

quality, spending, inter alia.2 In what follows, we argue that the advantage incentivizes

incumbents to largely ignore challengers, issues, and, even, candidates’ images. Instead, their

campaigns focus on their own experience, ties to the district/state, and their provision of benefits

for the district/state: factors on which incumbents are inherently advantaged. Voters, in turn,

focus on those criteria, putting aside policy and perceptions of candidates’ traits. The

consequence is minimal campaign engagement – which is so important for promissory

representation (e.g., Mansbridge, 2004; Druckman, 2014) and democratic competition more

1 While the incumbency advantage has shrunk in recent years – due to increased party loyalty and straight-ticket voting as well as the nationalization of congressional elections – incumbents still win re-election at remarkably high rates (Jacobson, 2015). In fact, in some sense, the fact that the incumbency advantage remains in the face of partisan sorting and nationalization makes it even more pressing to understand how the bias works (for recent work, see, e.g., Ashworth, Bueno de Mesquita, & Friedenberg, 2018; Fowler, 2018). 2 See, for example, Mann & Wolfinger (1980), Cox & Katz (2002), Prior (2006), Ashworth & Bueno de Mesquita (2008), Stone, Fulton, Maestas, & Maisel (2010), Eckles, Kam, Maestas, & Schaffner (2014), Hall (2014), Hall & Snyder (2015), and Fowler (2016). Other work studies representative communications and their effects (Grimmer 2013, 2016; also see Mayhew 1974; Fenno 1978), but not as it applies directly to the incumbency advantage.

3

generally (Garsten, 2009; Disch, 2011; c.f., Fowler, 2016). Furthermore, voters consequently

decide on criteria that have little connection to what one might consider evidence of substantive

representation. In essence, campaigns are a key mechanism through which the incumbency

advantage operates, and this is problematic for those concerned with campaign engagement and

democratic competition.

The Incumbency Advantage and Election Campaigns

The congressional incumbency advantage is one of the most widely studied topics in

American politics (Carson, Engstrom, & Roberts, 2007, 289). It reflects, in part, an incumbent’s

experience in office, familiarity (i.e., ties to the district), and the provision of benefits for the

district (e.g., casework, pork-barrel projects) (e.g., Mann & Wolfinger, 1980; Fiorina, 1989;

Gronke, 2001; Jacobson, 2013; Druckman, Kifer, & Parkin, 2009; Fowler & Hall, 2014, Fowler,

2018). These factors inherently favor the incumbent due to serving in the institution; holding

office itself is experience, and working on behalf of the district (i.e., taking actions for it) breeds

familiarity. Challengers lack the institutional access available to incumbents. This leads to a clear

hypothesis that voters view incumbents, relative to challengers, as possessing more experience

and familiarity, and as having taken more actions for the district (i.e., “incumbency factors”), all

else constant (hypothesis 1).

What this means for campaign rhetoric follows straightforwardly. Incumbents have an

advantage when it comes to experience, familiarity, and district actions, and thus most of their

campaign rhetoric will focus on those considerations (see Jacobson 2013, 109, 112). While not

identical, this is similar to Fenno’s (1978) homestyle, which entails explanation of Washington

activities (e.g., experience), presentation of self (e.g., familiarity), and allocation of resources

(e.g., district actions). We thus will refer to this strategy as “homestyle” – the use of the

4

homestyle strategy is meant to prime voters to rely on incumbency factors (experience,

familiarity, district actions) in their vote decisions.

The job for the challenger is to shift voters’ attention away from incumbency factors to

other criteria on which voters sometimes rely; this includes the proximity of the candidates’

policy positions to their own, and the candidates’ traits (i.e., perceived honesty, empathy, and

leadership) (Druckman, Kifer, & Parkin, 2009; Jacobson, 2013, 2015). Challengers can do this

by discussing the incumbent in a negative light; negativity is a way to stimulate attention, which

might be necessary to alert voters to move from their fallback reliance on incumbency features

(Marcus, Neuman, & MacKuen, 2000; Druckman & McDermott, 2008, Druckman, Kifer, &

Parkin 2010). Challengers also will emphasize issues and candidates’ images. In contrast to

incumbency factors, a challenger might be preferred when it comes to policy and/or traits (see

Jacobson, 2013, p. 104, 108); all else constant, those are the criteria on which challengers prefer

voters to rely. Challengers then will be more apt to use what we call an “issue/image” strategy.

How do these types of campaign rhetoric influence voters? Consider four premises. First,

as we intimated, unless particularly motivated, voters rely on incumbency criteria which are easy

to access and process (Mann & Wolfinger, 1980; Fowler, 2018; Ashworth et al. 2018). When

they are motivated to systematically process, voters will assess candidates’ policy positions and

traits, which take more cognitive resources to assess. Second, voters are more likely to engage in

systematic processing and consider policy and traits when they perceive there to be competition

(Chong & Druckman 2007, 640).3 Third, in congressional elections, active campaigns lead voters

to perceive increased competition (Bowler & Donovan 2011, 151; McDonald & Tolbert 2012).

And fourth, activity by the incumbent is the clearest signal of competition – this occurs when an

3 Competition signals to voters that their choices may matter, and so they invest more at arriving at their opinions. Evidence along these lines is that participation in elections increases with perceived competition (McDonald & Tolbert, 2012).

5

incumbent acknowledges the opponent, goes negative, and employs an issue/image strategy. Put

another away, voters fall back to incumbency factors unless they perceive a costly signal such as

an active campaign by the incumbent (e.g., Carter & Patty 2015, 836). An engaged campaign by

the challenger, on its own, is insufficient since voters recognize challengers have little choice but

to critique the incumbent’s record and image (Jacobson, 2013, p. 111). We thus hypothesize that

when an incumbent uses an issue/image strategy, voters are more likely to view the election as

competitive or close, all else constant (hypothesis 2). Moreover, when this happens, voters will

be more likely to base their votes on policy and candidate trait perceptions, all else constant

(hypothesis 3).

These hypotheses accentuate how campaign rhetoric can serve as a mechanism for the

incumbency bias. Incumbents hold disproportionate power since they enter campaigns

advantaged, given voters’ fallback is to rely on incumbency factors. And then, holding other

factors constant, it is only incumbent rhetoric that can prime voters to consider policy and

candidate traits. The result is an equilibrium of a disengaged campaign and voters ignoring

factors that drive substantive campaign dialogue (i.e., issues). As an aside, we recognize that

other forces such as poll results and media attention can alter voter decision-making; yet, we

focus strictly on the impact of campaign rhetoric so as to demonstrate its role, all else constant,

in perpetuating the incumbency advantage.

What Do Candidates Do?

Here we offer evidence that candidates campaign as we suggested: incumbents employ a

homestyle strategy of emphasizing familiarity, experience, and district actions, while challengers

go negative and emphasize issues/image. We follow Druckman et al. (2009) by offering a

content analysis of congressional candidate websites during the 2010 campaign. We focus on

6

2010 because, as will be clear below, our tests of the hypotheses come from a 2010

congressional campaign.

Websites, despite a relatively low number of visitors, provide an ideal measure of

campaign behavior by offering a holistic portrait of a campaign’s message aimed at general

voters.4 We drew a stratified random sample of major party House and Senate candidates; it was

stratified to ensure regional variation as well as some continuity with a larger over-time project.

The total sample included 369 sites.5 We hired a team of coders who participated in training and

practice, and then were assigned sets of sites to code. We provided detailed instructions for

coding a large array of features over the entire self-contained site. This included coding all parts

of the front-page, fundraising page, biography page, issues page, news page, and multi-media

page; coders did not follow links to other sites.

The sites were coded using measures we describe in the first column of Table 1. For

example, we measured issue campaigning in three ways: the number of partisan owned issues

discussed, the number of unambiguous positions taken, and the number of endorsements listed.

We include endorsements since that is a common method by which voters infer issue positions

(e.g., Lupia, 1994). Our image variable breaks out leadership and compassion/empathy while the

incumbency factors code for experience, familiarity, and district actions.

[Insert Table 1 About Here]

4 This is the case because campaigns take advantage of the infinite space on the web and they realize anything posted on their sites could potentially be passed along to voters in general (e.g., via the media). Further, Druckman et al. (2009, 2017) show, via surveys with campaign workers, that the websites encapsulate a candidate’s entire message and are aimed at the “average voter” in the district.5 We identified the population of candidates from the National Journal, Congressional Quarterly, and various state party home pages. We identified the central campaign website for each candidate via the National Journal’s website and/or Google searches. The websites coded were posted largely from mid-October to Election Day. The sample composition is 80% House sites, 20% Senate sites, 44% incumbent sites, 38% challenger sites, 18% open-seat candidate sites, 48% Democratic sites, and 52% Republican sites. Further, on the four-point Cook non-partisan competitiveness scale, 18% were toss-ups, 15% were likely, 12% were leaning, and 55% were solid. These percentages match the population of campaigns very well.

7

To assess the relative likelihood of incumbents and challengers employing each type of

strategy, we produced expected values on each variable. The values came from a set of

regressions (with control variables) presented in the Appendix. Specifically, we set other

variables at their mean values and then generated a predicted value for incumbents and then for

challengers. The results, reported in the second column of Table 1, make clear that challengers

are substantially more likely to employ an issue/image strategy, on all dimensions. All

differences between incumbents and challengers are statistically significant at the .05 level. For

example, virtually all challengers go negative – there is an average probability of .98 that they do

so as opposed to .62 for incumbents. Challengers are also substantially more likely to emphasize

partisan issues, take positions, and report endorsements. Challengers put more weight on image

as well – for instance, the probability that a challenger makes a leadership type statement is .56

compared to .34 for an incumbent. On the flip side, incumbents are dramatically more likely to

use a homestyle strategy by emphasizing experience, familiarity, and taking actions on behalf of

the district (e.g., they make, on average, four more statements about actions they have taken for

the district). Candidates do not engage in a dialogue, but rather focus on the criteria that likely

advantage them – and for incumbents that means simply emphasizing being an incumbent.

Experiment

Now that we have shown the approaches candidates employ, we turn to a test of our

predictions about the impact of these rhetorical strategies. We do this with an experiment that

allows us to control what messages voters receive and arrive at clear inferences about the impact

of those communications. We minimize the downside of such an approach by using real

candidates, which, in some sense, is critical to ensure respondents have prior opinions about the

incumbent. Specifically, we focus on the 2010 House election in Illinois’s 9th District that pitted

8

incumbent Democrat Jan Schakowsky against Republican challenger Joel Pollak. The district

mirrors the partisan lopsidedness of the bulk of districts in the United States (e.g., in 2012, 65%

of the ninth district voted for Obama) (Abramowitz, Alexander, & Gunning, 2006; Jacobson,

2013, p. 18). Yet, 2010 was the only year in decades with a bona fide challenger for the seat.

The increased expectation of competitiveness reflected, in part, Schakowsky’s brief

consideration of not pursuing re-election and instead running for the Senate. She also had some

personal controversy with her husband pleading guilty to tax withholding and bank fraud

charges. The 2010 challenger Joel Pollak had longstanding ties to the district and was an

outspoken advocate of Israel in a district with a sizable Jewish population. He also had received

national attention for a public exchange with Barney Frank and was endorsed by the Chicago

Tea Party. Even so, in the end Schakowsky beat Pollak 66% to 31%.

We recognize that our use of a female candidate running against a male candidate could

complicate matters, although the literature on voting and gender offers mixed evidence on

whether candidate gender affects voting behaviors (e.g., Brooks 2013). Regardless, one

possibility that we explore in the Appendix is whether explicit campaigning on leadership by

Schakowsky (when she uses an issue/image strategy) primes that trait since it is otherwise not

often associated with women candidates (e.g., Huddy & Terkildsen, 1993; Dolan, 2010). The

Appendix also offers additional details on the district and the 2010 election.6

Procedure and Design

6 We also looked at Pollak and Schakowsky’s actual 2010 websites. This confirmed that they generally followed the pattern found in our analysis of other 2010 congressional candidate websites. Although Schakowsky’s website included some negativity, Pollak’s site was awash in negative statements attacking the incumbent. Pollak also took clear positions on 31 issues (compared to 12 for Schakowsky), mentioned four endorsements (compared to two for Schakowsky), and highlighted leadership. The only strategy that Schakowsky employed more than Pollak was, predictably, mentioning six actions she had taken for the benefit of the district (whereas Pollak mentioned none).

9

We conducted the experiment from June to August 2010, which encapsulated the start of

the campaign but was prior to major campaign activity.7 We hired a professional website

designer to create Schakowsky and Pollak websites that drew content from the candidates’ own

webpages, candidate speeches, news coverage, and, for Schakowsky, floor votes. We then used

fliers and emails to recruit 395 participants from multiple colleges in the district and the

surrounding communities. Participants had to be eligible to vote in the district and received $10

compensation for their time. We provide a demographic portrait of the sample in the Appendix.

The study took place in a laboratory setting at pre-determined times. Upon arrival,

participants consented and then were provided with a packet. The packet explained that the

participant would read a brief overview of the ongoing U.S. congressional campaign, have 20

minutes to explore websites about the candidates, and then be asked to complete a survey. The

overview included: (1) a map of the district and pictures of the two candidates; (2) a statement

that many expected it to be the most competitive district race in some time; (3) an instruction that

the participant would have the next 20 minutes to explore websites for each candidate; and (4) an

instruction that time could be allocated in any manner, including not browsing and instead

reading magazines or newspapers we provided.

Each candidate site had a front-page with links to an issues page, a biography page, and,

for some conditions, an endorsement page. Our hypotheses require varying the candidates’

rhetorical strategies; we did this by randomly assigning participants to one of five conditions.

One condition served as a control, in which case the aforementioned directions were different,

with individuals spending time on non-campaign related websites and subsequently completing

the survey described below. The other conditions varied whether each candidate’s website

7 This ensured realism while at the same time minimizing the risk of strong pre-treatment effects (Druckman & Leeper, 2012) and/or incomparability of participants during the implementation period (e.g., due to major campaign events).

10

displayed factors consistent with the homestyle strategy or the issue/image strategy. In Table 2,

we show the details of each strategy, echoing the variables analyzed in our website coding (see

Table 1).8 For example, the issue/image strategy involved taking several clear issue positions,

offering endorsements, and discussing the candidate’s image; the homestyle strategy largely

avoided this information and instead focused on familiarity and actions for the district.9


In Table 3, we present a portrait of the four mixes of websites to which a given

respondent was (randomly) given access: Schakowsky issue/image X Pollak issue/image,

Schakowsky issue/image X Pollak homestyle, Schakowsky homestyle X Pollak issue/image, or

Schakowsky homestyle X Pollak homestyle.10 This mix of conditions allow us to see if campaign

strategy, all else constant, matters in the ways we predict.11


After browsing the websites for up to 20 minutes, respondents completed a survey that

asked about their perceptions of the candidates and election, as well as basic demographic and

political information. The survey included a question that asked how likely the respondent was to

vote for Pollak or Schakowsky on a 7-point scale ranging from “definitely will vote for Pollak”

8 We recognize that when it comes to issue positions and endorsements, the experiment includes fewer of each than is suggested by our web content analysis results. This reflects that many of the candidate web pages were much larger than the ones we could reasonably present in the experiment given time limitations. Also, we do not vary prior experience in office since Pollak had none and thus we could not realistically vary it.9 We extensively pre-tested the content to ensure our operationalization accurately captured the constructs described in Table 2 (e.g., positions were clearly taken or not, personal features involved the given category, etc.). The pre-test also gauged the leaning of the endorsers we used on the pages, showing that, for the most part, they were perceived in the correct partisan direction (i.e., Democratic for Schakowsky, Republican for Pollak). Details on the pre-test are in the Appendix.10 Our interest is mainly in conditions 2 through 5, where respondents are exposed to varying types of candidate rhetoric; for that reason, as well as practical considerations (e.g., resources), we collected a smaller sample for the control condition. 11 We opted to not vary competitiveness or the quality of the challenger so as to keep a clear focus on campaign rhetoric. Of course, all of these factors – quality, competitiveness, and rhetoric – affect one another. Our goal here is to establish the importance of rhetoric at which point future work can incorporate quality and competitiveness into a similar design.

11

to “definitely will vote for Schakowsky.” The survey also asked respondents to rate, on 7-point

scales, how well different traits including honesty, leadership, and compassion (empathy) fit

each candidate. We created relative trait assessment measures by subtracting the rating for Pollak

from the rating for Schakowsky for each item (and thus higher scores indicated relatively better

perceptions of Schakowsky; the theoretical range for the scale for each trait is -6 to 6).12 We

further asked participants to report their own issue position and each candidate’s position on 13

issues (all on 7-point scales). For each issue, we took the difference of the respondent’s position

from each candidate’s position, and then took the average difference (across all issues) for each

candidate. That provided us with scores that indicated how close the respondent felt his or her

issue positions, on average, were to each candidate.13 We then took the difference between these

two scores to arrive at an overall relative policy position proximity score such that higher scores

indicate greater perceived issue agreement with Schakowsky (the theoretical range for the scale

is -6 to 6).14

To gauge perceptions of incumbency features, we asked respondents to score which

candidate they believed possessed greater experience, greater familiarity, and had taken more

actions on behalf of voters in the district (on 7-point scales with higher scores moving towards

Schakowsky). We took the average across these items to arrive at an “incumbency” factor

12 The three traits scale together with an alpha of .60. This is not particularly high and thus in the Appendix we present results with each individual trait measure.13 For Pollak, the alpha is .85; for Schakowsky, the alpha is .80.14 For example, we measured the respondent’s issue position (on a 7-point scale) on whether the government should promote renewable energy sources. We also asked the respondent where he or she thought each candidate’s issue position fell on that issue. We took the difference between the respondent’s energy position and each perceived candidate position (the range then for this is 0 to 6). This gave us a score of “issue proximity” between the respondent and each candidate (i.e., proximity of energy position to Pollak and proximity of energy position to Schakowsky). We did this on all 13 issues and then took the average issue proximity score for each candidate (i.e., average closeness to Pollak across all issues and average closeness to Schakowsky across all issues). In this case, smaller scores indicate being closer to the given candidate on the issues. We then subtracted the overall Schakowsky score from the overall Pollak score (leading to a range of -6 to 6). Thus, 0 indicates indifference, positive scores indicate closeness to Schakowsky (i.e., the difference between the respondent and Schakowsky is smaller than the difference between the respondent and Pollak), and negative scores indicate closeness to Pollak.

12

score.15 We will use the incumbency factor score to test hypothesis 1 – that the incumbent is

always favored on these features, regardless of rhetoric. We then will regress the aforementioned

vote measure (which will then be our dependent variable) on our policy, trait, and incumbency

variables (which will be our independent variables) to test our hypothesis 3 that policy and traits

only should matter when the incumbent uses an issue/image strategy.

We also measured how close respondents thought the election would be on a 7-point

scale, from “not close at all” to “very close.” This allows us to test hypothesis 2 about perceived

closeness being driven by the incumbent’s use of the issue/image strategy. Finally, the survey

asked for participants’ partisan identification (with higher scores, on a 7-point scale, indicating

more Republican), gender, race/ethnicity (recoded to identify minority respondents), age

(offering 5 age ranges), income (offering 5 income ranges), and education (offering 5 levels of

highest education). All question wordings are in the Appendix. Participants were debriefed after

completing the survey.

Results

Our first hypothesis suggests that incumbents are inherently favored – regardless of the

campaign – on the incumbency features of experience, familiarity, and actions for the district.

We also predicted, in our second hypothesis, that when the incumbent employs an issue/image

strategy, it will signal to voters that the election is more competitive or close. We test these two

hypotheses, respectively, by reporting the mean scores by condition for our aggregated

incumbency feature variable and our measure of perceived election closeness. The results are in

Table 4. We see strong support for hypothesis 1: the incumbency factors are invariant to

campaign rhetoric. The mean scores not only strongly favor Schakowsky, with an overall

average of 5.20, but they do not significantly differ based on rhetoric. For instance, the lowest

15 The alpha is .80.

13

mean score is the control group with 5.05, and the highest is 5.27 (condition 5). These two do not

differ significantly (t136 = .84, p ≤ .40, for a two-tailed test).


This result accentuates just how privileged incumbents can be. Even when the challenger

highlights his experience, familiarity, and district actions, it does nothing to dislodge the inherent

preference for the incumbent on these dimensions. The incumbent’s advantage is further made

clear in the second column of Table 4 where we see strong support for hypothesis 2.

Respondents are significantly more likely to perceive the election as close only when the

incumbent Schakowsky employs an issue/image strategy. For example, in condition 2 where

both candidates use an issue/image strategy, the mean perceived closeness is 3.90. This is

substantially larger than the 3.27 score in condition 4 where Pollak highlights issues and image,

but Schakowsky does not (t170 = 3.52, p ≤ .01, for a two-tailed test). Overall, the combined mean

score for the incumbent issue/image conditions (i.e., conditions 2 and 3) is 3.90 while the

combined mean score for the other conditions is 3.22 (t391 = 5.73, p ≤ .01, for a two-tailed test). It

is the incumbent’s actions, regardless of what the challenger does, that signal a closer election.

We next explore the determinants of vote preference. Specifically, we regress the vote

preference variable on the incumbency factors, trait perceptions, and policy positions, along with

party identification, and other demographic variables. Recall the former four variables are all

coded such that higher values reflect movement towards Schakowsky. The first column of Table

5 shows that as an individual’s ratings for Schakowsky on incumbency, traits, and policy

increase, so does the likelihood that he or she will vote for Schakowsky. The negative coefficient

on partisanship shows that as one moves in a Republican direction, he or she becomes less likely

to vote for the Democrat Schakowsky. No other variables correlate with vote preference.

14


Recall that hypothesis 3 suggested that voters will be particularly likely to rely on traits

and policy only when the incumbent uses an image/issue strategy. We test this by adding

interactions between (1) a variable indicating experimental conditions where Schakowsky uses

an image/issue strategy (conditions 2 and 3) and (2) the traits, policy, and incumbency factor

variables. We also add analogous interactions for when Pollak uses an image/issue strategy – we

expect Pollak’s rhetoric will not matter here. The second column of Table 5 shows exactly that.

Specifically, we see highly significant interactions between the incumbent using an issue/image

strategy and voters considering policy and traits when thinking of their vote preference.

Moreover, the negative interaction with the incumbency factors variable reveals that when the

incumbent primes issues/image, voters move away from the very considerations on which the

incumbent is inherently favored. The set of non-significant interactions with the challenger

issue/image strategy make clear that it is the incumbent and not the challenger who has the

power to alter the nature of voter decision-making. This confirms hypothesis 3.16

To see the substantive impact of these rhetorical choices, consider a situation where

Pollak happens to be modestly favored by the electorate on both policy and traits; specifically,

assume on each of the scales that run from -6 to 6, with higher scores moving towards

Schakowsky, that the average voter locates at a -1. If Pollak employs an issue/image strategy but

Schakowsky goes homestyle, the average vote preference, on the 7-point scale, is 4.48 (standard

error = .25).17 – that is, a clear win for Schakowsky even though Pollak tries to prime issues and

16 It could be that respondents are projecting from their vote choice to perceived policy positions and traits (although see Hart & Middleton, 2014). However, even if they are doing that, the fact remains that the relationship between vote preference, traits, and policy is contingent on the incumbent’s strategy. It also is possible that instead of this occurring through priming, this process is more akin to learning (Lenz, 2013); from a theoretical standpoint, we are agnostic on this point since our focus is on which information voters are using and how the campaigns affect that. Finally, in the Appendix, we demonstrate that our results are robust to including additional interactions with different experimental conditions.17 We compute predicted values using Clarify (King, Tomz, & Wittenberg, 2000).

15

image. This dramatically changes if Schakowsky also employs an issue/image strategy with the

predicted vote average becoming 3.01 (.36): a 24.5% shift leading to Pollak having an edge.

Clearly, it is the incumbent’s strategy that matters.18

In sum, voters rely on the easily accessible and well-known incumbency factors unless

there is a signal to pay attention and process new trait and policy information (see Fowler, 2018;

Ashworth et al. 2018). That signal, at least when it comes to campaign rhetoric, is contingent on

the behavior of the incumbent using a strategy that is not in his/her interest to use.19 In short, the

campaign is a mechanism through which the incumbency advantage works. The factors that bias

voters towards the incumbent incentivize the incumbent to focus only on those criteria, and that

in turn, ensures the incumbency advantage. The challenger is in unenviable situation given

his/her campaign strategy cannot on its own induce voters to consider the criteria that could

advantage him/her.

Conclusion

Scholars and pundits have long worried about how inequities in campaign spending and

donations undermine democratic functioning. Part of this concern is that only well-resourced,

quality challengers could possibly mount an effective campaign against incumbents. In many

ways, our framework and results point to an equally troublesome dynamic. For us, resources are

held constant: the only variation concerns whether a candidate was an incumbent who used that

18 This example, though, is hypothetical. In actuality, respondents in our study preferred Schakowsky on both policy and traits, regardless of the experimental conditions. Put another way, Pollak failed to persuade voters even when they did rely on policy and traits in their decisions. In fact, in some cases, Pollak’s discussion of issues and images backfired and voters moved against him further. Details are in the Appendix. Our point, though, is that in the abstract Pollak could have been favored on these criteria, which is not the case when it comes to incumbency factors. 19The Table 5 results are robust if we instead use an ordered probit model. Also, recall that the alpha was relatively low for our trait variable. We thus – in the Appendix – re-run Table 5 Model 2 where we include the three individual trait measures (honesty, compassion, and leadership) instead of the single aggregate measure. We find that the trait effect displayed in Table 5 model 2 is driven by leadership. As mentioned in the text, this is interesting insofar as it may be that explicit attention to leadership primed it, as some work suggest it is otherwise counter-stereotypic (e.g., Huddy & Terkildsen, 1993; Dolan, 2009).

16

position to gain experience, establish familiarity, and take actions on behalf of the district. These

are all activities that, on their face, are unproblematic (Fowler, 2016). Yet, an inadvertent

consequence is that voters privilege incumbents. Voters’ limited attention and motivation makes

it extremely difficult for a challenger to launch a winning campaign. In our experiment, the

challenger would have had to rely on the incumbent making a miscalculation to even open the

door for voters to consider criteria on which he possibly could have competed.

Our study has clear limitations given the (experimental) focus on one particular campaign

in a district with a highly favored incumbent. We also did not attend to other factors that can

induce perceptions of competition. Our results, though, accentuate some key questions such as:

what other situational levers stimulate perceived competition?; when do they induce candidates

to campaign differently?; to what extent do campaigns, in varying scenarios, change electoral

outcomes?; and how much of the incumbency advantage comes from the campaign rhetoric, as

opposed to systematic or candidate specific reasons? Despite voluminous literatures on

congressional elections and campaign effects, these questions have gone largely unaddressed.20

Further, these questions connect to other studies of incumbency advantage. For example, are

potentially high-quality challengers deterred due to the anticipation of campaign dynamics that

disadvantage them? Zaller (1998) points to incumbents’ political skills as deterring challengers

from running (also see Levitt & Wolfram, 1997): is campaign ability that key political skill?

Our results also raise some important normative questions. Kelly (1960: 14) long ago

noted that campaigns ideally “expose the grounds on which candidates disagree and the

differences between the candidates – differences of personality, interest, affiliation, policy

20 Of particular relevance is the connection between how candidates choose strategies and their perception of competitiveness. That is, it may be that certain triggers stimulate incumbents to anticipate attentive voters and engage in an issue/image strategy to cater to the criteria on which voters may decide. That said, Enos and Hersh (2017) find that campaigns tend to be over-confident and that incumbents are more accurate in assessing electoral closeness

17

commitment, and all others that may affect performance in office.” Campaigns are “the main

point…of contact between officials and the populace over matters of public policy…

responsibility is imposed during campaigns and the elections in which they culminate” (Riker,

1996, 3; italics in the original). Our finding that incumbents do not engage and that voters rely on

non-substantive criteria makes even basic theories of retrospective accountability difficult to

satisfy (for discussion, see Druckman, 2014).

To be clear, it is certainly reasonable to expect that voters will choose high-quality

candidates (Ashworth & Bueno de Mesquita, 2008; Fowler, 2016) and that incumbents will win

re-election in districts with favorable partisan constituencies (Jacobson, 2015). After all,

Republican districts should naturally elect Republican lawmakers while Democratic districts

elect Democratic lawmakers. However, the problem for democratic theory comes from the fact

that voters are biased in favor of incumbent rhetoric. The rhetorical strategy commonly used by

incumbents may be less powerful in terms of rival party defections (Jacobson, 2015), but it

remains decisive in terms of supporters and independents, which ultimately results in easy re-

election in most cases. The key is that voters reward incumbents for doing next to nothing while

challengers are punished (or at least not rewarded) for trying to start the kind of substantive

debate that should be the center of all elections. Districts should get the representatives they want

but, ideally, the election outcome should be the result of a vigorous exchange of ideas – i.e., a

process in which campaign rhetoric leads voters to seriously consider their choices. There is

nothing normatively wrong with high-quality incumbents winning re-election, but there is

something wrong with their ability to win re-election by doing little more than reminding voters

that they are incumbents.

18

References

Abramowitz, A. I., Alexander, B., & Matthew Gunning. (2006). “Incumbency, Redistricting, and

the Decline of Competition in U.S. House Elections.” Journal of Politics, 68(1), 75-88.

Ansolabehere, S., Snyder, Jr., J. N., & Stewart III, C. (2000). “Old Voters, New Voters, and the

Personal Vote: Using Redistricting to Measure the Incumbency Advantage.” American

Journal of Political Science, 44(1), 17-34.

Ashworth, S., Bueno de Mesquita, E., & Friedenberg, A. (2018). “Learning about Voter

Rationality.” American Journal of Political Science, 62(1), 37-54.

Ashworth, S. & Bueno de Mesquita, E. (2008). “Electoral Selection, Strategic Challenger Entry,

and the Incumbency Advantage.” Journal of Politics, 70(4), 1006-1025.

Bowler, S. & Donovan, T. (2011). “Electoral Competition and the Voter.” Public Opinion

Quarterly, 75(1), 151–64.

Brooks, D.J. (2013). He Runs, She Runs: Why Gender Stereotypes Do Not Harm Women

Candidates. Princeton: Princeton University Press.

Carson, J. L. Engstrom, E. J., & Roberts, J. (2007). “Candidate Quality, the Personal Vote, and

the Incumbency Advantage in Congress.” American Political Science Review, 101(2),

289–301.

Carter, J., & Patty, J. W. (2015). “Valence and Campaigns.” American Journal of Political

Science, 59(4), 825–40.

Chong, D., & Druckman, J. N. (2007). “Framing Public Opinion in Competitive Democracies.”

American Political Science Review, 101(4), 637–55.

Cox, G. W., & Katz, J. N. (2002). Elbridge Gerry’s Salamander: The Electoral Consequences of

the Reapportionment Revolution. New York, NY: Cambridge University Press.

19

Disch, L. (2011). “Toward a Mobilization Conception of Democratic Representation.” American

Political Science Review, 105(1), 100–114.

Dolan, K. (2010). “The Impact of Gender Stereotypes Evaluations on Support for Women

Candidates.” Political Behavior, 32(1), 69-88.

Druckman, J. N. (2014). “Pathologies of Studying Public Opinion, Political Communication, and

Democratic Responsiveness.” Political Communication, 31(3), 467–92.

Druckman, J. N., Hennessy, C. L., Kifer, M. J., & Parkin, M. (2010). “Issue Engagement on

Congressional Candidate Web Sites, 2002—2006.” Social Science Computer Review 28

(1), 3–23.

Druckman, J. N., Kifer, M. J., & Parkin, M. (2009). “Campaign Communications in U.S.

Congressional Elections.” American Political Science Review, 103(3), 343–66.

Druckman, J. N., Kifer, M. J. & Parkin, M. (2010). "Timeless Strategy Meets New Medium:

Going Negative on Congressional Campaign Websites, 2002-2006." Political

Communication, 27(1), 88-103.

Druckman, J. N., Kifer, M. J. & Parkin, M. (2017). “Resisting the Opportunity for Change: How

Congressional Campaign Insiders Viewed and Used the Web in 2016.” Social Science

Computer Review, 36(4), 392-405.

Druckman, J. N., & Leeper, T. J. (2012). "Learning More from Political Communication

Experiments: Pretreatment and Its Effects." American Journal of Political Science, 56(4),

875-896.

Druckman, J. N., & McDermott, R. (2008). "Emotion and the Framing of Risky Choice."

Political Behavior, 30(3), 297-321.

20

Eckles, D. L., Kam, C.D, Maestas, C. L., & Schaffner, B. F., (2014). “Risk Attitudes and the

Incumbency Advantage.” Political Behavior, 36(4), 731–49.

Enos, R. D., & Hersh, E. D. (2017). "Campaign Perceptions, of Electoral Closeness: Uncertainty,

Fear and Over-Confidence." British Journal of Political Science, 47(3), 501–519.

Fenno, R. F. (1978). Home Style: House Members in Their Districts. Little, Brown.

Fiorina, M. P. (1989). Congress: Keystone of the Washington Establishment. New Haven, CT:

Yale University Press.

Fowler, A. (2018). "A Bayesian Explanation for the Effect of Incumbency." Electoral Studies,

53, 66-78.

Fowler, A. (2016). “What Explains Incumbent Success? Disentangling Selection on Party,

Selection on Candidate Characteristics, and Office-Holding Benefits.” Quarterly Journal

of Political Science, 11(3), 313–38.

Fowler, A., & Hall, A. B. (2014). “Disentangling the Personal and Partisan Incumbency

Advantages: Evidence from Close Elections and Term Limits.” SSRN Scholarly Paper ID

2120743. Rochester, NY: Social Science Research Network.

Garsten, B. (2009). “Representative Government and Popular Sovereignty.” In I. Shapiro, S. C.

Stokes, E. J. Wood, & A. S. Kirshner (Eds.) Political Representation (90-110). New

York, NY: Cambridge University Press.

Gelman, A., & King, G. (1990). "Estimating the Incumbency Advantage without Bias."

American Journal of Political Science, 34(4), 1142-64.

Grimmer, J. (2013). Representational Style in Congress: What Legislators Say and Why It

Matters. New York, NY: Cambridge University Press.

21

Grimmer, J. (2016). “Measuring Representational Style in the House: The Tea Party, Obama,

and Legislators’ ChangingExpressed Priorities.” In R. M. Alvarez (Ed.), Computational

Social Science: Discovery and Prediction (225-245). New York, NY: Cambridge

University Press.

Gronke, P. (2001). The Electorate, the Campaign, and the Office: A Unified Approach to Senate

and House Elections. Ann Arbor, MI: University of Michigan Press.

Hall, A.B. (2014). “Partisan Effects of Legislative Term Limits.” Legislative Studies Quarterly,

39(3), 407-429.

Hall, A.B., & Snyder, Jr., J.M. (2015). “How Much of the Incumbency Advantage is Due to

Scare-off?” Political Science Research and Methods, 3(3), 493–514.

Hart, A., & Middleton, J.A. (2014). “Priming under Fire: Reverse Causality and the Classic

Media Priming Hypothesis.” Journal of Politics, 76(2), 581–92.

Hayes, D. (2005). “Candidate Qualities through a Partisan Lens: A Theory of Trait Ownership.”

American Journal of Political Science 49 (4), 908–23.

Huddy, L., & Terkildsen, N. (1993). "Gender Stereotypes and the Perception of Male and

Female Candidates." American Journal of Political Science, 37(1), 119-147.

Jacobson, G. C. (2013). The Politics of Congressional Elections: Eighth Edition. Boston, MA:

Pearson.

Jacobson, G. C. (2015). "It's Nothing Personal: The Decline of the Incumbency Advantage in US

House Elections." Journal of Politics, 77(3), 861-873.

Kelley, Stanley Jr. (1960). Political Campaigning: Problems in Creating an Informed

Electorate. Washington, D.C.: The Brookings Institution.

22

King, G., Tomz, M., & Wittenberg, J. (2000). “Making the Most of Statistical Analyses:

Improving Interpretation and Presentation.” American Journal of Political Science, 44(2),

347–61.

Lee, D. S. (2008). “Randomized Experiments from Non-Random Selection in U.S. House

Elections.” Journal of Econometrics, 142(2), 675–97.

Lenz, Gabriel S. (2013). Follow the Leader?: How Voters Respond to Politicians’ Policies and

Performance. Chicago, IL: University of Chicago Press.

Levitt, S. D., & Wolfram, C. D. (1997). “Decomposing the Sources of Incumbency Advantage in

the U. S. House.” Legislative Studies Quarterly, 22(1), 45–60.

Lupia, A. (1994). "Shortcuts versus Encyclopedias: Information and Voting Behavior in

California Insurance Reform Elections." American Political Science rebiew, 88(1), 63-76.

Marcus, G. E., Neuman, W. R., & MacKuen, M. (2000). Affective Intelligence and Political

Judgment. Chicago, IL: University of Chicago Press.

Mann, T. E. & Wolfinger, R. (1980). "Candidates and Parties in Congressional Elections." The

American Political Science Review, 74(3), 617-632.

Mansbridge, J. (2004). “Representation Revisited: Introduction to the Case against Electoral

Accountability.” Democracy and Society, 2(1).

Mayhew, D. R. (1974). Congress: The Electoral Connection. New Haven, CT: Yale University

Press.

McDonald, M. P., & Tolbert, C. J. (2012). “Perceptions vs. Actual Exposure to Electoral

Competition and Effects on Political Participation.” Public Opinion Quarterly, 76(3),

538–54.

23

Prior, M. (2006). “The Incumbent in the Living Room: The Rise of Television and the

Incumbency Advantage in U.S. House Elections.” Journal of Politics, 68(3), 657–73.

Riker, William H. (1996). The Strategy of Rhetoric: Campaigning for the American

Constitution. Randall L. Calvert, John Mueller, and Rick K. Wilson (Eds.). New Haven,

CT: Yale University Press.

Stone, W. J., Fulton, S. A., Maestas, C. D., & Maisel, L. S. (2010). “Incumbency Reconsidered:

Prospects, Strategic Retirement, and Incumbent Quality in U.S. House Elections.”

Journal of Politics, 72(1), 178–90.

Zaller, John. (1998). “Politicians as Prize Fighters: Electoral Selection and Incumbency

Advantage.” In J. G. Geer (Ed.), Politicians and Party Politics (125–185). Baltimore,

MD: Johns Hopkins University Press.

24

Table 1: Candidate Website Features

Measure Predicted Values*

Issue/Image Strategy

Negativity Whether the candidate had any negativity (toward the opponent) on the site. (Probability of negativity.)

Incumbent: .62 (std. err.: .07).

Challenger: .98 (std. err.: .02)

IssuesWeighted relative public opinion partisan advantage of issues discussed on the site.A (Scale from -17.5% to 12.85%, is average public opinion advantage for candidate’s party on issues mentioned)

Number of unambiguous issue positions taken on the site (from 0 to 33).B

Number of endorsements on the site (from 0 to 311).C


Challenger: 1.98 (std. err.: .29)

Incumbent: 11.65 (std. err.: .58).

Challenger: 13.67 (std. err.: .76)

Incumbent: 10.85 (std. err.: 1.88).

Challenger: 16.68 (std. err.: 3.17)

Image D Whether the candidate made a statement about leadership (i.e., why running for office and the direction he or she will go if elected) on the front page or biography page. (Probability of leadership statement.)

Whether the candidate made a statement that signals compassion or empathy (i.e., details about his or her family) on the front page or biography page. (Probability of empathy statement.)





Homestyle StrategyE

Whether the candidate made a statement about experience (i.e., holding prior office) on the front page or biography page. (Probability of experience statement.)

Whether the candidate made a statement about being familiar with the district on the site (i.e., growing up or being from the district) on the front-page or biography page. (Probability of familiarity statement.)

Number of statements the candidate made about actions taken to benefit the district (e.g., including organizing events or working legislation, etc.) on the site (from 0 to 31).


Challenger: .25 (std. err.:.12)



Incumbent: 6.65 (std. err.: .71).

Challenger: 2.29 (std. err.: .28)*We computed predicted values using Clarify (King et al., 2000).A To compute this, we collected data from polls on the public’s perception of which party “owned” each issue (i.e., handled the issue better). For each issue mentioned on a candidate’s site, we took the average public opinion difference over all issues resulting in a partisan ownership variable with higher scores indicating increased ownership by the candidate’s party (see Hayes, 2005).B Audios and videos were not coded for this variable.C Audios and videos were not coded for this variable.D Statements about honesty were not coded. The operationalizations of leadership and compassion/empathy match those used in our experiment (also see Druckman et al., 2009 for rationale).E The operationalizations match those used in our experiment (also see Druckman et al., 2009 for rationale). Audios and videos were not coded for the action variable.

25

Table 2: Website Features

Issue/Image Strategy Homestyle Strategy NegativityA Negative statements about the

opponent (on the front-page and issues-page).

None.

Issues Four party-owned issues; one non-owned issue. (3 on front-page, 5 on issues page.) B

Three clear issue positions. (On issues page.)C

Thirteen endorsements.(On distinct endorsements page.) D

Five non-party-owned issues listed. (2 on front-page, 5 on issues page.)B

One clear issue position.(On issues page.)C

Five endorsements.(On issues page.)D

ImageE Statements about honesty, compassion/empathy, and leadership. (On front-page briefly and in detail on biography page.)

None.

HomestyleF None. Statements about familiarity, and actions taken on behalf of district. (On front-page briefly and in detail on biography page.)

A For example, on the issue/image Pollak site, the front-page included a statement that “It is time to replace Representative Jan Schakowsky; she is out of touch…,” and the issue-page included, “…Schakowsky cannot be trusted…” B We identified party owned issues based on public opinion data at the time; the five non-owned issues were taxes, deficit, immigration, morals/ethics, and government reform. The Democratic owned issues were health care, energy, education, and social security. The Republican owned issues were Homeland Security/terrorism, business, crime, and foreign policy/Middle East (Druckman, Hennessy, Kifer, & Parkin, 2010 for details on the construction of party ownership measures).C For Schakowsky issue/image strategy, the issue positions were on health care, energy, and education; for Schakowsky incumbent strategy, the issue position was on taxes. For Pollak issue/image strategy, the issue positions were on Homeland Security/terrorism, foreign policy/Middle East, and business. For Pollak incumbent strategy, the issue position was on government reform.D The endorsements came from well-known or fairly identifiable groups.E We operationalized variables as done in Druckman et al. (2009); honesty involved statements about trust; empathy involved family details; and leadership involved statements about running for office. We also included polls favorable to the given candidate for the issue/image (but not homestyle) sites (see Druckman et al., 2009).F We operationalized these as coded in Druckman et al. (2009); thus, familiarity involved statements about ties to district/history; and actions involved statements about participation in events, assistance, etc.

26

Table 3: Experimental Conditions

(1) = Control Group

Pollak Issue/Image Pollak Homestyle

Schakowsky Issue/Image (2) (3)

Schakowsky Homestyle (4) (5)

27

Table 4: Homestyle and Election Closeness Scores

Homestyle(1-7 scale, toward Sch.)

Election Closeness(1-7 scale, toward close)

1) Control 5.05(1.40; 44)

3.18(std. dev. = 1.13; N = 44)

2) Schakowsky Iss./Im. / Pollak Iss./Im.

5.18(1.22; 80)

3.90(1.17; 80)

3) Schakowsky Iss./Im. / Pollak Homestyle

5.20(1.17; 84)

3.89(1.21; 84)

4) Schakowsky Homestyle / Pollak Iss./Im.

5.23(1.34; 91)

3.27(1.17; 92)

5) Schakowsky Homestyle / Pollak Homestyle

5.27(1.46; 94)

3.19(1.09: 93)

Overall 5.20(1.31; 393)

3.50(1.20; 393)

Entries are averages with standard deviations and N in parentheses.

28

Table 5: Vote Preference Regressions

(Model 1) (Model 2)

Incumbency Factors 0.448*** 0.621***(0.054) (0.078)

Candidate Traits 0.177*** 0.119(0.056) (0.097)

Policy 0.384*** 0.225**(0.067) (0.101)

Party ID (Rep.) -0.158*** -0.117**(0.051) (0.049)

Female 0.038 0.069(0.137) (0.130)

Minority -0.154 0.042(0.160) (0.155)

Age -0.009 0.018(0.082) (0.078)

Income -0.045 -0.068(0.058) (0.055)

Education -0.020 0.043(0.092) (0.088)

Inc. Iss./Im. 1.672***(0.549)

Inc. Iss./Im. X Inc. Fac. -0.453***(0.106)

Inc. Iss./Im. X Traits 0.218**(0.106)

Inc. Iss./Im. X Policy 0.384***(0.114)

Chall. Iss./Im. -0.277(0.534)

Chall. Iss./Im. X Inc. -0.030Fac. (0.103)Chall. Iss./Im. X Traits -0.015

(0.106)Chall. Iss./Im. X Policy 0.083

(0.117)Constant 3.026*** 2.082***

(0.508) (0.562)

Observations 344 344R-squared 0.502 0.571Entries are OLS coefficients with standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 for two-tailed tests.

29

Appendix

Candidate Web Content Analysis Results

To produce the probabilities reported in Table 1, we ran a set of regressions on our outcome variables. We then produced predicted values using Clarify (Tomz et al., 2000). The regressions included a set of control variables, as follows. The results appear below in Tables A-1a, b.

The challenger and open seat variables are dummy variables (i.e., 0 or 1) indicating whether the candidate is a challenger or running for an open-seat. When both are equal to 0, it signifies incumbent status. Competition is a four-point score of how competitive the election was, based on the Cook non-partisan ratings. Senate, Democrat, and Female are dummy variables indicating whether the candidate is, respectively, running for the Senate, a Democrat, and a female. Funds raised is the amount of money the candidate raised according to the Federal Election Commissions. Front-runner is a three-point scale ranging from clear trailer to clear front-runner (see Druckman et al. 2009). District Republican is the percentage of the district that voted for the Republican presidential candidate in 2008. Opponent negativity is a dummy variable indicating whether the candidate’s opponent went negative on his or her website. Prior office is a dummy variable indicating whether the candidate held any prior elective office. We include that for the incumbency factor variables so as to differentiate pure incumbency advantage as opposed to general “prior office” advantage.

30

Table A-1a: Web Campaign FeaturesIssue

Negativity Ownership Positions EndorseChallenger 1.860*** 1.762*** 0.165* 0.439

(0.395) (0.455) (0.086) (0.307)Open Seat 1.117*** 0.678 0.218*** 0.535*

(0.315) (0.451) (0.081) (0.287)Competition 1.403*** -0.428 0.204*** 0.173

(0.344) (0.384) (0.070) (0.241)Senate -0.108 0.141 0.142* 0.436

(0.364) (0.448) (0.081) (0.276)Democrat 0.084 9.706*** -0.035 0.073

(0.215) (0.325) (0.060) (0.223)Female 0.154 0.316 -0.010 0.555**

(0.270) (0.387) (0.071) (0.251)Funds Raised 10.835** 2.499 0.218 1.066

(4.384) (2.093) (0.378) (1.357)Front-runner -0.490*** 1.045*** 0.017 0.055

(0.173) (0.235) (0.044) (0.149)District Republican 0.517 -1.649 -0.152 -1.264*

(0.788) (1.195) (0.229) (0.732)Opp. Negative 0.574**

(0.265)Constant -0.389 -5.918*** 2.401*** 2.510***

(0.707) (0.966) (0.184) (0.613)

Observations 369 367 369 369R-squared 0.749Log Likelihood -109.8 -881.1 -1196 -1256

Note: The negativity model is a probit regression; issue ownership is an ordinary least squares regression; and positions and endorsements are negative binomial regressions. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 for two-tailed tests.

31

Table A-1b: Web Campaign FeaturesDistrict

Leadership Empathy Experience Familiar Actions

Challenger 0.575*** 0.510** -1.098*** -0.498** -1.062***(0.219) (0.218) (0.294) (0.238) (0.193)

Open Seat 0.295 0.350* -0.740*** -0.276 -0.771***(0.215) (0.212) (0.283) (0.225) (0.170)

Competition -0.369** 0.335* 0.047 -0.085 0.259*(0.185) (0.182) (0.269) (0.192) (0.156)

Senate -0.251 0.092 -0.150 -0.377* 0.075(0.217) (0.211) (0.298) (0.218) (0.177)

Democrat 0.345** 0.273* 0.381* 0.100 0.584***(0.154) (0.153) (0.211) (0.152) (0.128)

Female -0.284 0.004 -0.291 -0.253 -0.154(0.184) (0.181) (0.249) (0.181) (0.150)

Funds Raised 1.937* -0.240 -1.970 -0.110 0.892(1.073) (0.987) (1.637) (1.070) (0.936)

Front-runner -0.038 0.092 -0.170 -0.131 -0.068(0.112) (0.112) (0.161) (0.117) (0.098)

District -0.816 0.191 0.533 0.258 0.476Republican (0.567) (0.559) (0.817) (0.557) (0.505)Prior Office 2.288*** 0.158 0.705***

(0.295) (0.209) (0.183)Constant -0.006 -0.601 -0.742 0.408 0.935**

(0.457) (0.457) (0.670) (0.463) (0.419)

Observations 369 369 369 369 369Log Likelihood -240.5 -245.1 -120.9 -245.6 -922.1

Note: The leadership, empathy, experience, and familiarity models are probit regressions; and district actions is a negative binomial regression. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 for two-tailed tests.

32

Details on the District/Campaign

Illinois’ 9th District includes northern parts of Chicago and northern suburbs and lies alongside Lake Michigan. During the 2010 campaign, district statistics, as listed at Congressional Quarterly’s Roll Call, stated an estimated population (from 2000) of 653,647, a median income of $46,531, and a demographic breakdown as follows: 68.5% White, 10.9% Black, 12.4% Asian, 11.5% Hispanic, 0.3% Native American, and 4.6% other. It is a highly Democratic district that had a Cook Partisan Voting Index Score of D+20 (meaning that the average Democratic share of the presidential vote in the district exceeds the national average share by 20%).

The District has been represented by a Democrat since 1949, with Jan Schakowsky being first elected in 1998. From 1998 to 2008, Schakowsky always garnered at least 70% of the vote. She has been one of the most liberal-leaning members of Congress. The 2010 election was expected to be at least relatively more competitive than prior contests (although it was still seen as a safe Democratic district). As mentioned in the text, Schakowsky ended up beating Pollak 66% to 31%.

33

Sample Demographics

Gender (n = 394) 60% femaleEthnicity/Race (n = 387) 70% White;

6% African-American; 15% Asian American; 4% Hispanic; 5% Other

Age (n = 391) 52% 18-24; 27% 25-34; 13% 35-50; 7% 51-65; 1% Over 65

Income (n = 363) 15% < $30,000; 35% $30,000-$69,999; 20% $70,000-$99,999; 21% $100,000-$200,000; 9% Over $200,000

Education (highest level) (n = 392) 4% High School; 37% Some College; 34% 4 Year College Degree; 25% Advanced Degree

Party Identification (n = 382) 25% Strong Democrat; 23% Weak Democrat; 22% Independent leans Democrat; 15% Independent; 7% Independent leans Republican; 5% Weak Republican; 3% Strong Republican

34

Pre-test

To ensure the webpage content was perceived as intended, we conducted a pre-test. To do this, we relied on a sample of 68 respondents who did not participate in the main experiment. We provided them with various segments meant to capture the variables (see Table 1) and asked them to rate them. For issue ownership, we listed a host of issues (e.g., see note below Table 2) and asked respondents which party they trust to do a better job at handling the issue (on a 7-point scale from “clearly Democrats” to “clearly Republicans”). For position-taking, we included statements and asked respondents to rate the extent to which it is an ambiguous or unambiguous in offering of a clear position (on a 7-point scale). For endorsements, we asked respondents to rate where each group political fell, from “strong Democrat” to “strong Republican” (on a 7-point scale). For each image, we asked respondents to rate the extent to which the statement portrayed the candidate as being a “leader,” “honest,” or “empathetic” (on 7-point scales). We used a similar approach for the incumbency factors of familiarity and taking actions. We present the results in the Table A-2, which provides mean value ratings across statements we used in the experiment. The results clearly suggest our content coheres with the constructs we intend.

35

Table A-2: Pre-Test ResultsDemocratic Issue Owned 2.54 (0.57)Republican Issue Owned 5.37 (0.33)No Issue Ownership 4.22 (0.46)Schakowsky Clear Position 6.10 (0.39)Pollak Clear Position 5.66 (0.75)Schakowsky Ambiguous Position 2.53 (0.49)Pollak Ambiguous Position 2.14 (0.30)Democratic Endorsements 2.45 (1.32)Republican Endorsements 5.05 (1.36)Schakowsky Leadership 5.67 (0.15)Pollak Leadership 5.61 (0.10)Schakowsky Honesty 5.27 (0.38)Pollak Honesty 5.13 (0.28)Schakowsky Empathy 4.50 (0.23)Pollak Empathy 5.28 (0.24)Schakowsky Familiarity 6.15 (0.37)Pollak Familiarity 6.10 (0.41)Schakowsky Actions 6.03 (0.15)Pollak Actions 5.92 (0.25)

Entries are averages with standard deviations in parentheses.

36

Experimental Survey Measures

How close do you think this election (Schakowsky vs. Pollak) will be?

1 2 3 4 5 6 7Not Close at

AllNot Sure Very Close

Using the scale below, please indicate how likely you are to vote for Pollak or Schakowsky. (If you plan to vote for neither, please circle 4.)

1 2 3 4 5 6 7Definitely

will vote for Pollak

Not Sure Definitely will vote for Schakowsky

Which candidate—Pollak or Schakowsky—do you think possesses greater experience in public office?


PollakNot Sure Definitely

Schakowsky

Which candidate—Pollak or Schakowsky—do you think is more familiar with the 9th Congressional District?



Schakowsky

Which candidate—Pollak or Schakowsky—do you think has taken more actions on behalf of voters in the 9th Congressional District?



Schakowsky

How well does each of the following words/phrases describe Pollak?

Not Well at

All

Not Sure

Very Well

Honest (e.g., trustworthy, forthcoming) 1 2 3 4 5 6 7

Compassionate 1 2 3 4 5 6 7

37

(e.g., cares about ordinary people, empathic)

Strong Leader (e.g., gets things done) 1 2 3 4 5 6 7

How well does each of the following words/phrases describe Schakowsky?

Not Well at

All

Not Sure

Very Well

Honest (e.g., trustworthy, forthcoming) 1 2 3 4 5 6 7

Compassionate (e.g., cares about ordinary people,empathic)

1 2 3 4 5 6 7

Strong Leader (e.g., gets things done) 1 2 3 4 5 6 7

Do you think the government should be less involved with regulating business/the economy (e.g., private industry, banking)?


NOTNot Sure Definitely

YES

Using the same 1-7 scale, where would you place Pollak’s position on business/economic regulation (i.e., what do

you think Pollak thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on business/economic regulation (i.e.,

what do you think Schakowsky thinks)?

Do you think there should be expanded health care coverage (e.g., requiring employers to provide coverage) in line with the recently passed health care law?



YES

Using the same 1-7 scale, where would you place Pollak’s position on health care (i.e., what do you think Pollak

thinks)?

38

Using the same 1-7 scale, where would you place Schakowsky’s position on health care (i.e., what do you think

Schakowsky thinks)?

Do you think the government should spend more on crime prevention (e.g., salaries for law-enforcement officers, community policing programs)?



YES

Using the same 1-7 scale, where would you place Pollak’s position on crime prevention spending (i.e., what do you

think Pollak thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on crime prevention spending (i.e., what

do you think Schakowsky thinks)?

Do you think the government should spend more on education (e.g., after-school programs, grants to modernize, renovate, and repair public schools)?



YES

Using the same 1-7 scale, where would you place Pollak’s position on education spending (i.e., what do you think

Pollak thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on education spending (i.e., what do you

think Schakowsky thinks)?

Do you think the government should lower capital gains taxes (i.e., taxes on the sale of stocks, bonds, and property)?



YES

Using the same 1-7 scale, where would you place Pollak’s position on capital gains taxes (i.e., what do you think

Pollak thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on capital gains taxes (i.e., what do you

think Schakowsky thinks)?

39

Do you think the government should be doing more to stop undocumented workers from entering the country?



YES

Using the same 1-7 scale, where would you place Pollak’s position on undocumented workers (i.e., what do you


Using the same 1-7 scale, where would you place Schakowsky’s position on undocumented workers (i.e., what do

you think Schakowsky thinks)?

Do you think the government should be more progressive (i.e., liberal) when it comes to moral and ethical issues like abortion and same-sex marriage?



YES

Using the same 1-7 scale, where would you place Pollak’s position on moral and ethical issues (i.e., what do you

think Pollak thinks)? _______________

Using the same 1-7 scale, where would you place Schakowsky’s position on moral and ethical issues (i.e., what do


Do you think the government should spend more on Social Security?



YES

Using the same 1-7 scale, where would you place Pollak’s position on Social Security spending (i.e., what do you


Using the same 1-7 scale, where would you place Schakowsky’s position on Social Security spending (i.e., what do


40

Do you think the government should be doing more to end government corruption and waste?



YES

Using the same 1-7 scale, where would you place Pollak’s position on government corruption (i.e., what do you


Using the same 1-7 scale, where would you place Schakowsky’s position on government corruption (i.e., what do


Do you think the government should take a tougher stand against Palestinian groups in the Middle East?



YES

Using the same 1-7 scale, where would you place Pollak’s position on the Middle East (i.e., what do you think

Pollak thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on the Middle East (i.e., what do you think

Schakowsky thinks)?

Do you think the government should reduce spending in order to lower the deficit?



YES

Using the same 1-7 scale, where would you place Pollak’s position on the deficit (i.e., what do you think Pollak

thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on the deficit (i.e., what do you think

Schakowsky thinks)?

41

Do you think the government should promote energy production from renewable sources (e.g., wind, biomass, geothermal, hydropower)?



YES

Using the same 1-7 scale, where would you place Pollak’s position on energy (i.e., what do you think Pollak

thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on energy (i.e., what do you think

Schakowsky thinks)?

Do you think the US government should sever relations (diplomatic, economic) with countries that harbor suspected terrorists?



YES

Using the same 1-7 scale, where would you place Pollak’s position on relations with countries that harbor suspected

terrorists (i.e., what do you think Pollak thinks)?

Using the same 1-7 scale, where would you place Schakowsky’s position on relations with countries that harbor

suspected terrorists (i.e., what do you think Schakowsky thinks)?

Are you male or female?

Male Female

Which of the following do you consider to be your primary racial or ethnic group?

White African American

Asian American Hispanic Native American

Other

What is your age?

18-24 25-34 35-50 51-65 Over 65

What is your estimate of your family’s annual household income (before taxes)?

< $30,000 $30,000 - $69,999 $70,000-$99,999 $100,000-$200,000 >$200,000

What is the highest level of education you have completed?

42

Less than High School High School Some College 4 Year College

Degree Advanced Degree

Generally speaking, do you consider yourself a Democrat, Independent, or Republican?

1 2 3 4 5 6 7Strong

DemocratWeak

DemocratIndependent

leans Democrat

Independent Independent leans

Republican

Weak Republican

Strong Republican

43

Supplementary Results/Models

In the first column of Table A-3, we replicate model 2 of Table 5 except we also include interactions for the conditions where both candidates used an issue/image strategy (i.e., condition 2) and where neither candidate used such a strategy (i.e., condition 5). We do this to confirm that it is specifically the incumbent actions that prime policy and traits, and to explore whether there is an added impact when both candidates use the issue/image strategy. The results show that the findings reported in Table 5 are largely robust. The interactions with incumbency factors and policy remain significant, although the one with traits falls just shy of statistical significance. Perhaps more importantly, no other interaction is significant, thereby confirming that it is the incumbent’s rhetorical choices, and only those choices, that influence voters’ criteria and it is his/her influence alone.

The second model in Table A-3 uses the individual items for the incumbency image (i.e., honesty, care, leadership) rather than the aggregate measure. We include this given that the aggregate measure does not have a particularly high alpha score. The results show that it is largely perception of leadership that drives the results. As noted in the text, this is interesting since it might be that explicit attention to leadership by a woman candidate primed it.

44

Table A-3: Vote Preference Regressions (1) (2)

VARIABLES All Interactions Trait InteractionsIncumbency Factors 0.757*** 0.648***

(0.143) (0.078)Candidate Traits 0.046 --

(0.199)Policy 0.030 0.198*

(0.293) (0.101)Party ID (Rep.) -0.105** -0.127***

(0.049) (0.049)Female 0.106 0.083

(0.131) (0.131)Minority -0.003 0.038

(0.157) (0.154)Age -0.007 0.035

(0.078) (0.079)Income -0.072 -0.060

(0.055) (0.055)Education 0.046 0.040

(0.088) (0.089)Inc. Iss./Im. 2.546*** 1.863***

(0.968) (0.551)Inc. Iss./Im. X Inc. Fac. -0.656*** -0.481***

(0.190) (0.107)Inc. Iss./Im. X Traits 0.309 --

(0.226)Inc. Iss./Im. X Policy 0.723** 0.415***

(0.320) (0.115)Chall. Iss./Im. 0.414 -0.095

(0.893) (0.540)Chall. Iss./Im. X Inc. -0.194 -0.065Fac. (0.173) (0.104)Chall. Iss./Im. X Traits 0.097 --

(0.223)Chall. Iss./Im. X Policy 0.304 0.092

(0.313) (0.118)Both Iss./Im. -1.383 --

(1.245)Both Iss./Im. X Inc. 0.258 --Fac. (0.243)Both Iss./Im. X Traits -0.134 --

(0.269)Both Iss./Im. X Policy -0.392 --

(0.355)None Iss./Im. -0.088 --

(0.935)None Iss./Im. X Inc. -0.078 --Fac. (0.179)None Iss./Im. X Traits 0.040 --

(0.242)None Iss./Im. X Policy 0.080 --

(0.317)Honesty -- 0.030

(0.080)Compassion -- 0.123*

(0.064)Leadership -- -0.076

(0.066)Inc. Iss./Im. X Honesty -- 0.054

(0.088)Inc. Iss./Im. X -- -0.088Compassion (0.069)Inc. Iss./Im. X -- 0.294***Leadership (0.075)Chall. Iss./Im. X -- -0.007Honesty (0.091)Chall. Iss./Im. X -- -0.007Compassion (0.075)Chall. Iss./Im. X -- 0.017Leadership (0.077)Constant 1.637* 1.874***

(0.856) (0.566)Observations 344 341R-squared 0.589 0.590

Entries are OLS coefficients with standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 for two-tailed tests.

45

In Table A-4, we present the mean scores for overall vote choice, policy positions, and trait perceptions. The most notable result is the lack of variance on vote choice – Pollak’s chances do not improve when voters attend to policy and traits. In fact, there are no significant differences across conditions and in all cases, Schakowsky is the clear favorite (e.g., comparing conditions 1 and 5, gives t135 = .77, p ≤ .45, for a two-tailed test).

To understand why this is the case, consider the final two columns of the table. When Schakowsky uses an issue/image strategy, leading voters to consider policy and traits, she also effectively persuades them to favor her to a greater extent on image. Even when Pollak introduces traits (condition 2), Schakowsky still persuades such that respondents are significantly more favorable towards her. The merged means on traits for when Schakowsky uses an issue/image strategy is 1.03 versus .46 when she does not (t390 = 3.92, p ≤ .01, for a two-tailed test). Unlike the incumbency factors, image perceptions are moveable via campaign rhetoric but it is the incumbent Schakowsky who is persuasive and thus she wins even when using an issue/image strategy.

Likewise, issue perceptions can be affected by campaign rhetoric but Pollak is unable to leverage this into an advantage. In the cases where he explicitly discusses his issue positions (i.e., when he employs an issue/image strategy, conditions 2 and 4), it backfires as respondents move further in favor of Schakowsky. Consider condition 3 where positions matter to voters but Pollak opts for a homestyle strategy and does not discuss issues. In that case, the average perceived policy proximity score is .36. Condition 2 differs only in that Pollak discusses his issue positions and the score actually increases toward Schakowsky to 1.02 (t162 = 3.06, p ≤ .01, for a two-tailed test).

Table A-4: Vote, Image, and Issue Proximity Scores

Vote(1-7 scale, toward Sch.)

Policy(-6 to 6 scale, toward Sch.)

Traits(-6 to 6 scale, toward Sch.)

1) Control 5.13(1.34; 45)

0.35(0.82; 41)

0.31(1.11; 44)

2) Schakowsky Iss./Im. / Pollak Iss./Im.

4.98(1.90; 80)

1.02(1.56; 80)

1.21(1.47; 80)

3) Schakowsky Iss./Im. / Pollak Homestyle

5.02(1.57; 82)

0.36(1.22; 84)

0.86(1.40; 84)

4) Schakowsky Homestyle / Pollak Iss./Im.

5.04(1.72; 91)

0.73(1.49; 92)

0.57(1.77; 91)

5) Schakowsky Homestyle / Pollak Homestyle

4.91(1.72; 92)

0.03(1.31; 92)

0.42(1.16; 93)

Overall 5.00(1.68; 390)

0.50(1.39; 389)

0.70(1.46; 392)

Entries are averages with standard deviations and n in parentheses.

46